## wpiea2021065-print-pdf - REFERENCES

## Source details

**Canonical URL:** [wpiea2021065-print-pdf - REFERENCES](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021065-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021065-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021065-print-pdf.pdf.json)

---

### I. Introduction and contribution to the literature
- COVID-19 declared a global pandemic in early March 2020; the “Great Lockdown” is characterized as a supply and demand shock.
- Paper’s empirical focus:
  - Examines determinants of countries’ fiscal measures in response to COVID-19, emphasizing the role of speed and type of containment measures.
  - Regresses governments’ fiscal measures on Public Health Response Time (PHRT: speed of introducing stringency measures) and controls.
  - Differentiates “hard/physical” lockdowns and “soft/smart” measures (testing, contact tracing, public health information campaigns).
  - Controls: average stringency, income, fiscal space, budget transparency, economic outlook, health preparedness.
  - Sample period: daily data on stringency indices between January 1st and October 15th, 2020.
  - Fiscal measures data: covers 190 countries from the IMF October 2020 Fiscal Monitor, by type (above-the-line and below-the-line).
- Main high-level finding:
  - Faster and smarter containment measures were associated with lower fiscal responses.
  - Swift introduction of (especially smart) measures may reduce or nullify the need to maintain on average higher restrictions over time.

### II. Data sources, indices, and PHRT definition
- OxCGRT indices used:
  - "stringency index" (school closures; workplace closures; public event cancellations; gathering restrictions; public transportation closures; stay-at-home orders; restrictions on internal movement; international travel bans).
  - "containment and health index" (all of the above plus testing policy, contact tracing and public health information campaigns).
- Sample period: Jan. 1 – Oct. 15, 2020.
- PHRT defined as: the number of days it takes a country to reach its maximum stringency (measured by both the OxCGRT stringency and containment and health indices) after a significant outbreak.
  - Significant outbreak defined as: 100 confirmed cases.
  - PHRT measures days to each country’s maximum index over the sample period.
- Empirical notes:
  - A country reaching maximum before 100 cases occurs in about 21 percent of the sample.
  - Cases where two or more countries had about the same stringency at 100 cases occurred collectively in about 70 percent of the sample.

### III. Stylized facts and regional/income patterns
- By income level:
  - Reported fiscal responses highest in AEs, followed by EMs and LIDCs.
  - Health responses smaller than non-health and other measures, on average.
  - Additional spending on health as a share of overall fiscal response was most sizable in LIDCs.
  - BTL-CL measures more pronounced in AEs.
- By region:
  - Median announced fiscal response highest in EUR, at around 10 percent of GDP.
  - APD and WHD follow EUR.
  - Median fiscal response in MCD and AFR countries was relatively the smallest.
- PHRT patterns:
  - AFR countries were the fastest to reach maximum restrictiveness (using both indices).
  - WHD region exhibited the largest dispersion of PHRT, especially using the containment and health index.
  - All regions were slower to enact health and testing responses compared to lockdowns (median for containment and health index is uniformly higher than median for stringency index).
  - EUR median PHRT improves relative to other regions, suggesting more aggressive testing policies.
- Anecdotal relationship:
  - Visual examination suggests faster PHRT associated with lower fiscal responses (percent of GDP).
  - Countries’ income levels appear relevant to fiscal response size.

### IV. Empirical approach and model specification
- Cross-sectional regressions with dependent variable = fiscal response (share of GDP).
  - Basic specification uses total fiscal response (ATL+BTL+CL); separate specifications for ATL health and BTL-CL.
- Core regression: fiscalresponse_i = β0 + β1 phrt_i + β2 avg_i + β3 lgdppc_i + β4 X_i + ε
- Key policy variables:
  - phrt: PHRT measured using OxCGRT stringency index (captures physical lockdowns).
  - phrt_smart: PHRT measured using OxCGRT containment and health index (captures testing/contact tracing/public health).
  - avg, avg_smart: average OxCGRT daily stringency and containment and health indices over sample period.
- Controls included:
  - (log of) GDP per capita in real PPP terms (October 2020 IMF WEO).
  - Fiscal space proxy: (log of) Emerging Markets Bond Index (EMBI) spreads.
  - Budget transparency: 2019 Open Budget Survey (OBS) index.
  - Health variables: reported cases per population (OxCGRT), population ages 65+ share (WDI), 2019 Global Health Security (GHS) index.
- Empirical caveats:
  - Results show associations, not causal inference.
  - Dependent variable uses announced fiscal measures (may differ from implementation).
  - OxCGRT does not capture voluntary social distancing or gaps between official and actual implementation.

### V. Main empirical results (preserving reported coefficients exactly)
- Core finding: faster and smarter containment associated with smaller fiscal responses.
  - phrt_smart coefficient positive and statistically significant across specifications for total fiscal response (models 1–3), ATL health (models 7–9), and BTL-CL (models 13–14).
  - phrt (physical lockdowns only) coefficient not statistically different from zero in 6 of 8 specifications in tables 1–3.
- Magnitude summaries reported:
  - Simple average of statistically significant phrt_smart coefficients in tables 1–4 is around +0.03, indicating that a 10 percent decrease in mean PHRT is associated with a lower fiscal stimulus by 0.3 percent of GDP, holding all else constant.
  - Simple average of statistically significant avg and avg_smart coefficients in tables 1–4 is around -0.08, indicating that a 10 percent increase in mean stringency is associated with a lower fiscal stimulus by 0.8 percent of GDP, holding all else constant.
- Selected coefficients preserved exactly as reported:
  - Table 1 highlights:
    - phrt_smart: 0.016* (model 1), 0.028** (model 2), 0.046*** (model 3).
    - avg_smart: -0.102* (model 1), -0.112* (model 2), -0.040 (model 3).
    - phrt: 0.016 (model 4), 0.020 (model 5), 0.031*** (model 6).
    - avg: -0.097** (model 4), -0.093* (model 5), 0.001 (model 6).
    - lgdppc_rppp: 2.948*** (model 1), 2.738*** (model 2), -1.299 (model 3), 2.920*** (model 4), 2.852*** (model 5), -0.619 (model 6).
    - obsi2019: 0.086*** (model 1), 0.052* (model 2), 0.080** (model 4).
    - lembi: -1.422** (model 5).
    - Observations: 147 (models 1 and 4), 105 (models 2 and 5), 56 (models 3 and 6).
    - R-squared: 0.292 (model 1), 0.383 (model 2), 0.426 (model 3).
  - Table 2 (Health Fiscal Response):
    - phrt_smart: 0.002* (models 7, 8, 9).
    - avg_smart: -0.007; -0.010**; -0.012 (models 7–9).
    - casespop: 0.236***; 0.200**; 0.245***; 0.189** (models 7–12).
    - Observations: 128; 128; 58; 128; 128; 58.
    - R-squared: 0.045; 0.140; 0.152; 0.015; 0.117; 0.112.
  - Table 3 (BTL and Contingent Liability Measures):
    - phrt_smart: 0.028***; 0.031** (models 13, 14).
    - avg_smart: -0.062; -0.014 (models 13, 14).
    - phrt: 0.027**; 0.008 (models 15, 16).
    - obsi2019: 0.088***; 0.024; 0.086***; 0.016 (models 13–16).
    - lembi: -0.538; -0.411 (models 13–16).
    - Observations: 106; 56; 106; 56.
    - R-squared: 0.179; 0.305; 0.163; 0.121.
  - Table 4 (Robustness Checks):
    - phrt_smart: 0.011* (model 17, ATL_nonhealth).
    - phrt_smart: 0.054*** (model 18, Total_deff).
    - phrt_smart: 0.056*** (model 19, Total).
    - phrt_smart: 0.025** (model 20, Total).
    - phrt_smart: 0.018** (model 21, Total).
    - avg_smart: -0.008; -0.009; -0.050; -0.016; -0.082 (models 17–21).
    - lgdppc_rppp: -0.689; -0.860; -1.230; -0.153; 2.901*** (models 17–21).
    - casespop: 1.013**; 1.361 (models 17 and 19).
    - lembi: -0.811***; -1.546**; -1.567**; -0.958* (models 17–20).
    - revggw: -0.249* (model 20).
    - GGdebt: 0.085*** (model 21).
    - GGdebtlidc: -0.082*** (model 21).
    - GGfiscbal: 0.306** (model 21).
    - lidc: 5.683** (model 21).
    - Observations: 52; 56; 56; 60; 143.
    - R-squared: 0.341; 0.421; 0.404; 0.206; 0.416.

### VI. Robustness, initial conditions, and interpretation
- Robustness:
  - phrt_smart coefficient remains positive and statistically significant across multiple alternative specifications and dependent variable measures.
  - avg_smart coefficient generally insignificant, consistent with the notion that early strict interventions (smart and fast) reduced the need for prolonged high restrictions.
- Fiscal space and initial conditions:
  - EMBI spreads (lembi) negative and significant for ATL non-health (model 17), indicating fiscal space binding for non-health measures.
  - Alternative fiscal space measures:
    - GGdebt positive and significant (model 21).
    - GGdebtlidc negative and significant (model 21).
    - GGfiscbal positive and significant (model 21).
  - revggw negative and significant (model 20), implying worsening economic outlook prompted larger overall fiscal response.
- Sample statistics:
  - mean of phrt_smart ≈ 34 days; 1 st dev ≈ 54 days.
  - mean of avg_smart ≈ 48 (index 0–100).
  - Note: "A 10 percent increase in the mean is thus equivalent to around ½ st dev."

### VII. Policy implications and conclusions
- Smart and rapid containment measures can reduce both health and economic losses and reduce the size of fiscal responses to shocks like COVID-19.
- Fiscal space matters particularly for non-health fiscal measures and in developing countries; access to finance and initial fiscal conditions shaped response size and design.
- Higher budget transparency and stronger institutional and PFM capacity are associated with larger and more active BTL-CL fiscal measures.
- The same measures that help save lives (fast and smart containment) can also save fiscal resources.

*Source: wpiea2021065-print-pdf - REFERENCES _____________________________________________________________________________________________21*

### REFERENCES _____________________________________________________________________________________________21

### wpiea2021065-print-pdf - REFERENCES _____________________________________________________________________________________________21

### I. Introduction and contribution to the literature
- COVID-19 is a health and economic shock; declared a global pandemic in early March 2020.
- The “Great Lockdown” is characterized as a supply and demand shock disrupting production and affecting households and businesses.
- Policymakers faced trade-offs in balancing health and economic responses.
- Observational and empirical evidence cited:
  - Cowling et al (2020): NPIs effective in reducing incidence in Hong Kong between January and March 2020.
  - IMF (2020), Deb et al (2020a), Demirgüç-Kunt et al (2020): NPIs effective, especially when implemented faster.
- SIR-based and related model literature:
  - Berger et al (2020): expanding testing with targeted quarantine dampens economic fallout and reduces peak symptomatic infections.
  - Acemoglu et al (2020), Brotherhood et al (2020), Checo et al (2020): targeted policies by age and risk groups effective.
  - Piguillem and Shi (2020): random testing can substitute for quarantines.
  - Forslid and Herzing (2020): early quarantine postpones but does not alter infections; cost increases with duration and extent of quarantine.
  - Lattanzio and Palumbo (2020): importance of soft containment measures (social distancing, masks, sanitizing).
  - Andrabi et al (2020): importance of smart real-time testing, contact tracing, community messaging.
  - Cherif and Hasanov (2020): universal testing strategy requires sacrificing test accuracy for scalability and industrial policy to ramp up tests.
- High-frequency indicator literature:
  - Proxies: electricity usage, nitrogen dioxide emissions, mobility trends, job postings (Chen et al (2020); Deb et al (2020a; 2020b); Demirgüç-Kunt et al (2020); IMF (2020)).
  - Debt et al (2020b): containment measures had, on average, an impact equivalent to a loss of about 15 percent in industrial production.
  - Deb et al (2020a) and Demirgüç-Kunt et al (2020): early NPIs limit economic fallout.
  - Cevik and Öztürkkal (2020): more infections associated with higher sovereign CDS spreads.

- This paper’s empirical focus:
  - Examines determinants of countries’ fiscal measures in response to COVID-19, focusing on role of speed and type of containment measures.
  - Regresses governments’ fiscal measures on observed public health response time (PHRT: speed of introducing stringency measures) and other controls.
  - Differentiates “hard/physical” lockdowns and “soft/smart” measures (testing, contact tracing, public health information campaigns).
  - Controls include average stringency over sample period, income, fiscal space, budget transparency, economic outlook, health preparedness.
  - Sample period: daily data on stringency indices between January 1st and October 15th.
  - Fiscal measures data: covers 190 countries, by type (above-the-line and below-the-line) from the IMF October 2020 Fiscal Monitor.

- Main finding:
  - Faster and smarter containment measures were associated with lower fiscal responses.
  - Faster PHRT is better; containment type matters—physical lockdowns alone are less effective than when accompanied by smart measures.
  - Swift introduction of (especially smart) measures may reduce or nullify need to maintain on average higher restrictions over time.
  - Suggests no necessary trade-off between health and economic outcomes; same measures that save lives can save fiscal resources.

- Additional findings on initial conditions:
  - Fiscal space constrained overall fiscal response, especially non-health measures and in developing countries.
  - Fiscal packages larger for higher-income countries, especially below-the-line measures.
  - Lower-income countries spent more on health given weaker initial health infrastructure and preparedness.
  - Health-related variables (cases per population, share of elderly population) matter for health fiscal response.
  - Higher budget transparency associated with larger fiscal response, reflecting importance of institutional and PFM capacity.

- Contribution relative to related empirical work:
  - Differentiates physical and smart containment measures, emphasizes speed of enacting stringency, and studies impact on government fiscal responses.
  - Uses updated dataset (Jan 1st till Oct 15th) and controls for more variables compared to Fotiou and Lagerborg (2021a/b) and Balajee et al (2020).

### II. An initial look at the data (stylized facts)
- Data source: October 2020 Fiscal Monitor database on country fiscal measures in response to COVID-19.
- Dataset classification:
  - On-budget above-the-line (ATL) health and non-health measures, tax deferrals.
  - Off-budget below-the-line (BTL) and contingent liabilities (CLs such as guarantees and quasi-fiscal operations).
  - ATL includes forgone revenues and additional spending (support to households).
  - BTL-CL measures mostly support firms.

- Key stylized facts:
  - By income level:
    - Reported fiscal responses highest in AEs, followed by EMs and LIDCs.
    - Health responses smaller than non-health and other measures, on average.
    - Additional spending on health as a share of overall fiscal response was most sizable in LIDCs.
    - BTL-CL measures more pronounced in AEs.
  - By region:
    - Median announced fiscal response highest in European (EUR) countries, at around 10 percent of GDP.
    - Asian-Pacific (APD) and Western Hemisphere (WHD) follow EUR.
    - Median fiscal response in Middle East and Central Asia (MCD) and African (AFR) countries was relatively the smallest, potentially reflecting weaker fiscal space going into the crisis.

### III. Empirical approach and structure of the paper
- Empirical strategy:
  - Regress fiscal measures on PHRT and other controls to identify determinants of fiscal responses.
  - Distinguish early/smart containment measures and capture dynamics of governments’ stringency responses.
- Paper organization:
  - Section II: stylized facts.
  - Section III: empirical model and results.
  - Section IV: conclusion.

*Source: wpiea2021065-print-pdf - REFERENCES _____________________________________________________________________________________________21*

### 19.  This is also reflected in  the IMF Policy  Tracker, available at https://www.imf.org/en/Topics/imf-and-

### wpiea2021065-print-pdf - 19.  This is also reflected in  the IMF Policy  Tracker, available at https://www.imf.org/en/Topics/imf-and-

### Data sources and indices
- OxCGRT indices used:
  - "stringency index" (school closures; workplace closures; public event cancellations; gathering restrictions; public transportation closures; stay-at-home orders; restrictions on internal movement; international travel bans).
  - "containment and health index" (all of the above plus testing policy, contact tracing and public health information campaigns).
- Sample period: Jan. 1 – Oct. 15, 2020.
- OxCGRT data reference: Hale et al 2020; Oxford’s COVID-19 Government Response Tracker dataset.

### Definition and measurement of Public Health Response Time (PHRT)
- PHRT defined as: the number of days it takes a country to reach its maximum stringency (measured by both the OxCGRT stringency and containment and health indices) after a significant outbreak.
- Significant outbreak defined as: 100 confirmed cases (consistent with Deb et al (2020a), Fotiou and Lagerborg (2021a; 2021b), and Baldwin and di Mauro (2020)).
- Distinctions from prior work:
  - PHRT measures days to each country’s maximum index over the sample period (not just days to increase index after outbreak, nor just the index value at 100 cases).
  - International travel restrictions are retained in the stringency measure.
  - The containment and health index is used as-is (not reconstructed to separate health component).
- Empirical note: the example where a country reaches maximum before 100 cases (country A) occurs in about 21 percent of the sample; cases where two or more countries had about the same stringency at 100 cases (countries B and C example) occurred collectively in about 70 percent of the sample.

### Regional patterns of PHRT and containment
- AFR countries were the fastest to reach maximum restrictiveness (using both OxCGRT stringency and containment and health indices).
- WHD region exhibited the largest dispersion of PHRT, especially using the containment and health index.
- All regions were slower to enact health and testing responses compared to lockdowns (median for containment and health index is uniformly higher than median for stringency index).
- EUR median PHRT improves relative to other regions, suggesting more aggressive testing policies.

### Anecdotal relationship between speed of containment and fiscal response
- Visual examination suggests: faster PHRT associated with lower fiscal responses (in percent of GDP).
- Countries’ income levels appear relevant to fiscal response size.
- The paper intends to empirically test these hypotheses controlling for initial conditions and other variables.

### Empirical model specification
- Cross-sectional regression: dependent variable = fiscal response (share of GDP). Basic specification uses total fiscal response (ATL+BTL+CL); separate specifications for ATL health and BTL-CL.
- Core regression:
  fiscalresponse_i = β0 + β1 phrt_i + β2 avg_i + β3 lgdppc_i + β4 X_i + ε
- Key policy variables:
  - phrt: PHRT measured using OxCGRT stringency index (captures physical lockdowns).
  - phrt_smart: PHRT measured using OxCGRT containment and health index (captures testing/contact tracing/public health).
  - avg, avg_smart: average OxCGRT daily stringency and containment and health indices over sample period.
- Expectations:
  - β1 expected positive when using PHRT expressed as days (faster containment = lower PHRT) implying faster containment (lower PHRT) associated with smaller fiscal responses.
  - avg could be positive (higher lockdowns costly) or avg_smart could be negative (higher smart measures less costly).

### Control variables included
- (log of) GDP per capita in real PPP terms (October 2020 IMF WEO).
- Fiscal space proxy: (log of) Emerging Markets Bond Index (EMBI) spreads.
- Budget transparency: 2019 Open Budget Survey (OBS) index.
- Health variables: reported cases per population (OxCGRT), population ages 65+ as share of total (WDI), 2019 Global Health Security (GHS) index.
- Note: panel methodology not used because time-series for dependent variable (fiscal responses) is not available.

### Empirical caveats
- Causality is difficult: containment decisions depend on pandemic evolution, which affects fiscal responses and vice versa. Reported regression results show associations, not causal inference.
- Dependent variable uses announced fiscal measures (may differ from implementation; may be staged; mixes discretionary measures and automatic stabilizers).
- Coverage differences in government levels may blur ATL vs BTL-CL distinctions.
- Late-hit countries may have learned from others, affecting PHRT.
- OxCGRT does not capture voluntary social distancing or gaps between official and actual implementation.

### Main empirical results and interpretation
- Core finding: faster and smarter containment associated with smaller fiscal responses.
  - The coefficient on phrt_smart is positive and statistically significant across reported specifications for total fiscal response (models 1-3), ATL health (models 7-9), and BTL-CL (models 13-14).
  - When using phrt (physical lockdowns only), coefficient not statistically different from zero in 6 of 8 specifications in tables 1-3.
- Magnitude summary:
  - Simple average of statistically significant phrt_smart coefficients in tables 1-4 is around +0.03, indicating that a 10 percent decrease in mean PHRT is associated with a lower fiscal stimulus by 0.3 percent of GDP, holding all else constant.
  - Simple average of statistically significant avg and avg_smart coefficients in tables 1-4 is around -0.08, indicating that a 10 percent increase in mean stringency is associated with a lower fiscal stimulus by 0.8 percent of GDP, holding all else constant.
- Interpretation: importance of acting fast and acting smart (testing/contact tracing) — consistent with modeling and empirical literature showing testing and tracking reduce infections and disrupt economy less than blanket lockdowns.

### Selected model estimates (Table 1 highlights)
- phrt_smart coefficients reported: 0.016* (model 1), 0.028** (model 2), 0.046*** (model 3). (Standard errors shown in table.)
- avg_smart coefficients reported: -0.102* (model 1), -0.112* (model 2), -0.040 (model 3).
- phrt coefficients reported: 0.016 (model 4), 0.020 (model 5), 0.031*** (model 6).
- avg coefficients reported: -0.097** (model 4), -0.093* (model 5), 0.001 (model 6).
- lgdppc_rppp coefficients reported: 2.948*** (model 1), 2.738*** (model 2), -1.299 (model 3), 2.920*** (model 4), 2.852*** (model 5), -0.619 (model 6).
- obsi2019 coefficients reported: 0.086*** (model 1), 0.052* (model 2), 0.080** (model 4).
- lembi coefficient reported: -1.422** (model 5).
- Observations: 147 (models 1 and 4), 105 (models 2 and 5), 56 (models 3 and 6).
- R-squared examples: 0.292 (model 1), 0.383 (model 2), 0.426 (model 3).

### Notable sample statistics
- In sample: mean of phrt_smart ≈ 34 days; 1 st dev ≈ 54 days.
- In sample: mean of avg_smart ≈ 48 (index goes from 0-100).

*Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021065-print-pdf.pdf*

### 10.  A 10 percent increase in  the mean is thus  equivalent to around ½  st dev.

### 10.  A 10 percent increase in  the mean is thus  equivalent to around ½  st dev.

### Key findings
- The paper’s main finding: faster and smarter containment measures are associated with lower fiscal responses to COVID-19.
- Fiscal space constrained the overall fiscal response, with the estimated coefficient for EMBI spreads negative.
- Fiscal packages were larger for higher-income countries, especially BTL-CL measures; lower-income countries spent more on health fiscal measures.
- Health-related variables mattered for the health fiscal response: higher cases per population are associated with higher health fiscal responses.
- Higher budget transparency is associated with larger fiscal measures, especially BTL-CL measures.

### Evidence from regressions (selected coefficients preserved exactly as reported)
- Table 2 (Determinants of the Health Fiscal Response):
  - phrt_smart: 0.002* (models 7, 8, 9)
  - avg_smart: -0.007; -0.010**; -0.012 (models 7–9)
  - lgdppc_rppp: -0.039; -0.149**; -0.103; -0.037; -0.157*; -0.062 (models 7–12)
  - GHSI: -0.002; -0.002; -0.005; -0.000; -0.002; -0.007 (models 7–12)
  - casespop: 0.236***; 0.200**; 0.245***; 0.189** (models 7–12)
  - Popages65: 0.013; 0.010; 0.013; 0.014 (models 7–12)
  - lembi: -0.052; -0.030 (models 7–12)
  - Observations: 128; 128; 58; 128; 128; 58
  - R-squared: 0.045; 0.140; 0.152; 0.015; 0.117; 0.112

- Table 3 (Determinants of BTL and Contingent Liability (CL) Measures):
  - phrt_smart: 0.028***; 0.031** (models 13, 14)
  - avg_smart: -0.062; -0.014 (models 13, 14)
  - phrt: 0.027**; 0.008 (models 15, 16)
  - obsi2019: 0.088***; 0.024; 0.086***; 0.016 (models 13–16)
  - lembi: -0.538; -0.411 (models 13–16)
  - Observations: 106; 56; 106; 56
  - R-squared: 0.179; 0.305; 0.163; 0.121

- Table 4 (Robustness Checks):
  - phrt_smart: 0.011* (model 17, ATL_nonhealth)
  - phrt_smart: 0.054*** (model 18, Total_deff)
  - phrt_smart: 0.056*** (model 19, Total)
  - phrt_smart: 0.025** (model 20, Total)
  - phrt_smart: 0.018** (model 21, Total)
  - avg_smart: -0.008; -0.009; -0.050; -0.016; -0.082 (models 17–21)
  - lgdppc_rppp: -0.689; -0.860; -1.230; -0.153; 2.901*** (models 17–21)
  - obsi2019: 0.016; 0.054*; 0.047 (models 17–19)
  - casespop: 1.013**; 1.361 (models 17 and 19)
  - lembi: -0.811***; -1.546**; -1.567**; -0.958* (models 17–20)
  - revggw: -0.249* (model 20)
  - GGdebt: 0.085*** (model 21)
  - GGdebtlidc: -0.082*** (model 21)
  - GGfiscbal: 0.306** (model 21)
  - lidc: 5.683** (model 21)
  - Observations: 52; 56; 56; 60; 143
  - R-squared: 0.341; 0.421; 0.404; 0.206; 0.416

### Robustness and specification notes
- The phrt_smart coefficient remains positive and statistically significant across multiple alternative specifications and dependent variable measures.
- The avg_smart coefficient is generally insignificant, suggesting early strict interventions (smart and fast) reduced the need for prolonged high restrictions.
- EMBI spreads (lembi) are negative and statistically significant for ATL non-health (model 17), indicating fiscal space was binding especially for ATL non-health measures.
- Alternative measures of fiscal space (GGdebt, GGdebtlidc, GGfiscbal) indicate:
  - GGdebt positive and significant (countries with higher debt had larger fiscal packages, potentially reflecting AEs capacity to carry debt).
  - GGdebtlidc negative and significant (higher GG debt in developing countries constrained the fiscal response).
  - GGfiscbal positive and significant (stronger initial fiscal balance facilitated larger stimulus).
- Revision to real GDP growth (revggw), computed as January 2020 WEO versus October 2020 WEO, has a negative and significant coefficient, implying worsening economic outlook prompted a larger overall fiscal response.

### Policy implications and conclusions
- Smart and rapid containment measures can reduce both health and economic losses and also reduce the size of fiscal responses to shocks like COVID-19.
- Fiscal space matters particularly for non-health fiscal measures and in developing countries; access to finance and initial fiscal conditions shaped response size and design.
- Higher budget transparency and stronger institutional and PFM capacity are associated with larger and more active BTL-CL fiscal measures.
- The same measures that help save lives (fast and smart containment) can also save fiscal resources.

*Source: wpiea2021065-print-pdf - 10.  A 10 percent increase in  the mean is thus  equivalent to around ½  st dev.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021065-print-pdf.pdf_